Intermarket_VV0 Trading_NN1 Strategies_NN2 Optimization_NN1 Trading_NN1 system_NN1 optimization_NN1 is_VBZ a_AT1 controversial_JJ subject_NN1 among_II analysts_NN2 and_CC traders_NN2 ._. 
But_CCB what_DDQ is_VBZ optimization_NN1 ?_? 
When_CS designing_VVG a_AT1 technical_JJ trading_NN1 system_NN1 ,_, the_AT analyst_NN1 must_VM use_VVI an_AT1 indicator_NN1 or_CC formula_NN1 to_TO produce_VVI trading_NN1 signals_NN2 ._. 
All_DB indicators_NN2 ,_, however_RR ,_, contain_VV0 one_MC1 or_CC more_DAR variable_JJ parameters_NN2 ._. 
A_AT1 parameter_NN1 can_VM be_VBI set_VVN to_II one_MC1 of_IO several_DA2 possible_JJ numeric_JJ values_NN2 ,_, which_DDQ can_VM greatly_RR alter_VVI the_AT behavior_NN1 and_CC performance_NN1 of_IO the_AT indicator_NN1 ._. 
The_AT dilemma_NN1 is_VBZ what_DDQ value_NN1 to_TO assign_VVI to_II each_DD1 parameter_NN1 ._. 
That_DD1 is_VBZ where_RRQ optimization_NN1 occurs_VVZ ._. 
Optimization_NN1 is_VBZ the_AT process_NN1 of_IO choosing_VVG the_AT best_JJT parameter_NN1 values_NN2 for_IF the_AT indicator_NN1 ._. 
This_DD1 is_VBZ accomplished_VVN by_II having_VHG the_AT software_NN1 "_" try_VV0 "_" a_AT1 large_JJ number_NN1 of_IO different_JJ combinations_NN2 of_IO parameters_NN2 or_CC combination_NN1 of_IO indicators_NN2 ,_, until_CS a_AT1 set_NN1 is_VBZ found_VVN that_CST gives_VVZ the_AT best_JJT result_NN1 ._. 
The_AT result_NN1 could_VM be_VBI net_JJ profits_NN2 ,_, percentage_NN1 of_IO profitable_JJ trades_NN2 ,_, minimum_JJ drawdown_NN1 and_RR31 so_RR32 forth_RR33 ._. 
An_AT1 example_NN1 of_IO a_AT1 trading_NN1 system_NN1 parameter_NN1 is_VBZ the_AT length_NN1 of_IO a_AT1 moving_JJ average_NN1 used_VVD ._. 
It_PPH1 is_VBZ very_RG important_JJ to_TO identify_VVI the_AT ranges_NN2 into_II which_DDQ the_AT values_NN2 for_IF parameters_NN2 and_CC criteria_NN2 fall_VV0 ._. 
A_AT1 thorough_JJ development_NN1 of_IO a_AT1 trading_NN1 system_NN1 must_VM involve_VVI the_AT testing_NN1 of_IO thousands_NNO2 of_IO parameter_NN1 combinations_NN2 to_TO identify_VVI those_DD2 that_CST persist_VV0 in_II producing_VVG profits_NN2 ,_, even_CS21 though_CS22 they_PPHS2 are_VBR not_XX necessarily_RR the_AT optimum_JJ ._. 
Thus_RR ,_, it_PPH1 is_VBZ often_RR better_JJR to_TO pick_VVI up_RP those_DD2 parameters_NN2 in_II the_AT best_JJT ten_MC percent_NNU of_IO the_AT parameters_NN2 instead_II21 of_II22 the_AT best_RRT based_VVN on_II stable_JJ neighbouring_JJ parameters_NN2 ._. 
Today_RT 's_GE software_NN1 for_IF trading_VVG systems_NN2 can_VM take_VVI you_PPY down_II a_AT1 dangerous_JJ path_NN1 ._. 
A_AT1 big_JJ profit_NN1 ,_, however_RR ,_, on_II an_AT1 optimized_JJ test_NN1 does_VDZ not_XX necessarily_RR mean_VVI you_PPY have_VH0 a_AT1 good_JJ system_NN1 ._. 
The_AT danger_NN1 with_IW optimization_NN1 is_VBZ that_DD1 what_DDQ worked_VVD well_RR in_II the_AT past_NN1 will_VM almost_RR certainly_RR not_XX perform_VVI as_RR21 well_RR22 in_II the_AT future_NN1 ._. 
Optimized_JJ variables_NN2 can_VM reflect_VVI irregularities_NN2 in_II the_AT data_NN far_RG more_DAR than_CSN any_DD underlying_JJ logic_NN1 of_IO the_AT market_NN1 or_CC the_AT trading_NN1 strategy_NN1 ._. 
Usually_RR ,_, such_DA curve-fitting_NN1 fails_VVZ as_CSA soon_RR as_CSA the_AT parameters_NN2 hit_VVD the_AT real_JJ world_NN1 with_IW its_APPGE random_JJ shocks_NN2 ,_, because_CS the_AT more_RRR control_VV0 you_PPY take_VV0 of_IO the_AT data_NN by_II specifying_VVG indicators_NN2 and_CC their_APPGE optimal_JJ parameters_NN2 ,_, the_AT less_DAR substance_NN1 is_VBZ left_VVN in_II the_AT data_NN to_TO provide_VVI predictive_JJ capacity_NN1 ._. 
Statisticians_NN2 call_VV0 this_DD1 a_AT1 loss_NN1 of_IO degrees_NN2 of_IO freedom_NN1 ;_; traders_NN2 call_VV0 it_PPH1 curve-fitting_JJ ._. 
Some_DD analysts_NN2 claim_VV0 that_CST if_CS the_AT data_NN sample_NN1 is_VBZ large_JJ enough_RR ,_, then_RT the_AT curve-fitting_NN1 captures_VVZ most_DAT of_IO the_AT price_NN1 dynamics_NN and_CC the_AT future_NN1 buy_VV0 and_CC sell_VV0 signals_NN2 will_VM be_VBI just_RR as_RG good_JJ as_CSA the_AT curve-fitted_JJ ones_NN2 ._. 
These_DD2 claims_NN2 are_VBR generally_RR accepted_VVN as_CSA being_VBG true_JJ because_CS there_EX is_VBZ a_AT1 lack_NN1 of_IO knowledge_NN1 ,_, but_CCB the_AT truth_NN1 is_VBZ that_CST they_PPHS2 are_VBR based_VVN on_II nothing_PN1 more_DAR than_CSN hope_NN1 ._. 
So_RR how_RRQ can_VM we_PPIS2 design_VVI trading_NN1 models_NN2 and_CC avoid_VVI the_AT illusion_NN1 of_IO optimization_NN1 ?_? 
The_AT usual_JJ way_NN1 to_TO validate_VVI an_AT1 optimized_JJ trading_NN1 system_NN1 is_VBZ to_TO test_VVI the_AT system_NN1 on_II data_NN it_PPH1 has_VHZ never_RR been_VBN tested_VVN on_RP before_II --_NN1 that_REX21 is_REX22 ,_, test_VV0 the_AT system_NN1 on_II some_DD data_NN set_VV0 aside_RL and_CC not_XX used_VVN when_CS developing_VVG the_AT model_NN1 ._. 
This_DD1 holdout_NN1 data_NN set_NN1 is_VBZ referred_VVN to_II as_RG out-of-sample_JJ data_NN ._. 
If_CS the_AT system_NN1 performance_NN1 holds_VVZ up_RP well_RR on_II the_AT out-of-sample_JJ data_NN ,_, we_PPIS2 can_VM be_VBI confident_JJ the_AT system_NN1 will_VM perform_VVI as_RR21 well_RR22 in_II the_AT future_NN1 as_CSA it_PPH1 has_VHZ in_II the_AT past_NN1 ._. 
A_AT1 more_RGR rigorous_JJ method_NN1 of_IO back-testing_NN1 is_VBZ that_DD1 of_IO walk_NN1 forward_RL testing_VVG ._. 
The_AT concept_NN1 is_VBZ similar_JJ to_TO out-of-sample_VVI testing_NN1 but_CCB instead_II21 of_II22 optimizing_VVG on_II seven_MC years_NNT2 of_IO data_NN and_CC using_VVG the_AT last_MD three_MC years_NNT2 for_IF testing_NN1 ,_, a_AT1 system_NN1 is_VBZ optimized_VVN on_II five_MC years_NNT2 and_CC then_RT tested_VVD the_AT next_MD year_NNT1 ._. 
Once_RR this_DD1 test_NN1 is_VBZ completed_VVN ,_, the_AT whole_JJ time_NNT1 window_NN1 is_VBZ moved_VVN forward_RL one_MC1 year_NNT1 to_TO include_VVI the_AT year_NNT1 just_RR tested_VVN and_CC the_AT system_NN1 is_VBZ then_RT re-optimized_VVN ._. 
It_PPH1 is_VBZ then_RT tested_VVN again_RT the_AT next_MD year_NNT1 ._. 
This_DD1 process_NN1 is_VBZ repeated_VVN again_RT and_CC again_RT ,_, "_" walking_VVG forward_RL "_" through_II the_AT data_NN series_NN until_CS all_DB the_AT data_NN is_VBZ exhausted_VVN ._. 
At_II the_AT end_NN1 ,_, all_DB tabulated_JJ one-year_JJ out-of-sample_JJ results_NN2 are_VBR merged_VVN to_TO create_VVI one_MC1 large_JJ out-of-sample_JJ results_NN2 segment_NN1 and_CC the_AT system_NN1 performance_NN1 is_VBZ evaluated_VVN based_VVN on_II the_AT combined_JJ out-of-sample_JJ results_NN2 ._. 
The_AT main_JJ advantage_NN1 of_IO using_VVG this_DD1 method_NN1 is_VBZ that_CST since_CS market_NN1 dynamics_NN change_VV0 slowly_RR over_II time_NNT1 the_AT optimum_JJ values_NN2 based_VVN on_II the_AT most_RGT recent_JJ five-year_JJ segment_NN1 will_VM be_VBI accurate_JJ in_II forecasting_VVG one_MC1 year_NNT1 into_II the_AT future_NN1 before_II the_AT model_NN1 falls_VVZ apart_RL ._. 
The_AT time_NNT1 periods_NN2 used_VVN in_II this_DD1 example_NN1 for_IF optimization_NN1 and_CC walk_VV0 forward_RL testing_VVG are_VBR not_XX set_VVN in_II stone_NN1 ._. 
The_AT selection_NN1 of_IO the_AT optimization_NN1 period_NN1 depends_VVZ on_II the_AT available_JJ data_NN and_CC the_AT time_NNT1 required_VVN to_TO model_VVI the_AT market_NN1 dynamics_NN ._. 
The_AT walk_NN1 forward_RL period_NN1 can_VM be_VBI between_II 12-20_MCMC %_NNU of_IO the_AT optimization_NN1 period_NN1 ._. 
Although_CS computationally_RR intensive_JJ ,_, this_DD1 is_VBZ an_AT1 excellent_JJ way_NN1 to_TO study_VVI and_CC test_VVI a_AT1 trading_NN1 system_NN1 ._. 
Even_CS21 though_CS22 optimization_NN1 is_VBZ occurring_VVG ,_, in_II a_AT1 sense_NN1 all_DB the_AT trades_NN2 are_VBR tested_VVN on_II what_DDQ is_VBZ essentially_RR out-of-sample_JJ test_NN1 data_NN ._. 
In_RR21 addition_RR22 ,_, the_AT results_NN2 will_VM closely_RR simulate_VVI the_AT process_NN1 that_CST occurs_VVZ during_II real_JJ trading_NN1 ,_, in_II which_DDQ traders_NN2 frequently_RR re-optimize_VV0 their_APPGE systems_NN2 to_TO bring_VVI them_PPHO2 up_JJ31 to_JJ32 date_JJ33 with_IW fundamental_JJ or_CC technical_JJ changes_NN2 in_II the_AT traded_JJ market_NN1 or_CC its_APPGE intermarket_NN1 relationships_NN2 ._. 
Even_CS21 though_CS22 nothing_PN1 can_VM guarantee_VVI future_JJ results_NN2 ,_, this_DD1 approach_NN1 appears_VVZ to_TO be_VBI as_RG close_JJ as_CSA we_PPIS2 can_VM practically_RR come_VVI to_II estimating_VVG trading_NN1 relationships_NN2 during_II constantly_RR changing_JJ market_NN1 conditions_NN2 ._. 
Unfortunately_RR ,_, totally_RR automated_JJ walk_NN1 forward_RL testing_VVG is_VBZ not_XX available_JJ in_II most_DAT popular_JJ off-the-shelf_JJ trading_NN1 software_NN1 and_CC implementing_VVG it_PPH1 manually_RR can_VM be_VBI time_NNT1 consuming_VVG ._. 
There_EX is_VBZ ,_, however_RR ,_, a_AT1 Tradestation_NN1 add-in_NN1 that_CST can_VM be_VBI purchased_VVN separately_RR ._. 
Walk_VV0 forward_RL testing_VVG is_VBZ also_RR available_JJ in_II Deepinsight_NP1 Professional_JJ and_CC TradersStudio_NP1 ._. 
A_AT1 major_JJ drawback_NN1 of_IO out-of-sample_JJ testing_NN1 is_VBZ that_CST to_TO exhaustively_RR determine_VVI the_AT optimal_JJ indicator_NN1 sets_VVZ as_II31 well_II32 as_II33 their_APPGE respective_JJ periods_NN2 would_VM require_VVI an_AT1 inordinately_RR large_JJ amount_NN1 of_IO data_NN ._. 
If_CS one_PN1 uses_VVZ this_DD1 approach_NN1 consistently_RR ,_, every_AT1 piece_NN1 of_IO data_NN must_VM be_VBI thrown_VVN away_RL after_II a_AT1 single_JJ use_NN1 in_II the_AT testing_NN1 phase_NN1 ._. 
In_II31 addition_II32 to_II33 the_AT cost_NN1 of_IO subscribing_VVG to_II data_NN providers_NN2 ,_, some_DD futures_NN2 have_VH0 only_RR recently_RR begun_VVN trading_NN1 ._. 
If_CS the_AT amount_NN1 of_IO data_NN you_PPY possess_VV0 is_VBZ limited_VVN ,_, you_PPY may_VM want_VVI to_TO consider_VVI the_AT following_JJ additional_JJ methods_NN2 of_IO reliability_NN1 testing_NN1 ._. 
Cluster_VV0 spotting_VVG and_CC profit_NN1 mapping_NN1 :_: When_CS optimizing_VVG a_AT1 system_NN1 ,_, traders_NN2 often_RR save_VV0 the_AT parameter_NN1 set_NN1 which_DDQ produces_VVZ the_AT highest_JJT level_NN1 of_IO profit_NN1 ._. 
Once_CS the_AT system_NN1 with_IW the_AT highest_JJT level_NN1 of_IO profit_NN1 is_VBZ identified_VVN ,_, you_PPY should_VM inspect_VVI the_AT results_NN2 of_IO surrounding_JJ parameter_NN1 values_NN2 nearby_RL to_TO be_VBI sure_JJ it_PPH1 is_VBZ not_XX merely_RR the_AT result_NN1 of_IO a_AT1 freak_JJ coincidence_NN1 in_II the_AT data_NN ._. 
Optimal_JJ results_NN2 which_DDQ are_VBR surrounded_VVN by_II similar_JJ levels_NN2 of_IO profit_NN1 are_VBR much_RR more_RGR desirable_JJ ._. 
A_AT1 better_JJR approach_NN1 is_VBZ to_TO export_VVI the_AT test_NN1 results_NN2 produced_VVN from_II each_DD1 simulation_NN1 to_II an_AT1 Excel_VV0 spreadsheet_NN1 and_CC plot_VV0 a_AT1 three-dimensional_JJ graph_NN1 ,_, with_IW one_MC1 parameter_NN1 scaled_VVN along_II the_AT X-axis_NN1 ,_, the_AT second_MD parameter_NN1 scaled_VVN along_II the_AT Y-axis_NN1 and_CC the_AT profit_NN1 as_CSA height_NN1 along_II the_AT Z-axis_NN1 ._. 
Casually_RR inspecting_VVG this_DD1 graph_NN1 ,_, one_PN1 can_VM find_VVI the_AT region_NN1 where_RRQ variations_NN2 in_II both_DB2 parameters_NN2 have_VH0 the_AT smallest_JJT impact_NN1 on_II profitability_NN1 ._. 
Parameter_NN1 sets_VVZ in_II this_DD1 range_NN1 may_VM be_VBI assumed_VVN to_TO be_VBI more_RGR reliable_JJ than_CSN parameter_NN1 sets_NN2 which_DDQ produce_VV0 spikes_NN2 of_IO profits_NN2 ._. 
This_DD1 is_VBZ because_CS low_JJ parameter_NN1 sensitivity_NN1 in_II the_AT past_NN1 implies_VVZ continued_JJ low_JJ sensitivity_NN1 in_II the_AT future_NN1 ,_, even_CS21 though_CS22 the_AT parameter_NN1 values_NN2 producing_VVG maximum_JJ profits_NN2 are_VBR outside_II this_DD1 parameter_NN1 range_NN1 ._. 
The_AT process_NN1 described_VVN above_RL can_VM be_VBI applied_VVN to_II a_AT1 wide_JJ variety_NN1 of_IO trading_NN1 systems_NN2 as_CSA most_DAT of_IO them_PPHO2 are_VBR either_RR already_RR described_VVN by_II two_MC parameters_NN2 or_CC can_VM be_VBI altered_VVN to_TO fit_VVI a_AT1 two-parameter_JJ model_NN1 by_II optimizing_VVG each_DD1 rule_NN1 or_CC indicator_NN1 separately_RR ._. 
Monte_NP1 Carlo_NP1 simulation_NN1 :_: This_DD1 technique_NN1 is_VBZ mainly_RR used_VVN to_TO obtain_VVI a_AT1 more_RGR realistic_JJ figure_NN1 for_IF the_AT drawdown_NN1 statistic_NN1 than_CSN the_AT figure_NN1 obtained_VVN by_II simulation_NN1 testing_NN1 and_CC is_VBZ performed_VVN by_II rearranging_VVG the_AT trades_NN2 randomly_RR (_( hence_RR the_AT name_NN1 )_) ._. 
To_TO rearrange_VVI the_AT trades_NN2 ,_, the_AT Monte_NP1 Carlo_NP1 simulation_NN1 uses_VVZ a_AT1 random_JJ number_NN1 generator_NN1 to_TO select_VVI a_AT1 random_JJ number_NN1 between_II zero_MC and_CC one_MC1 ._. 
This_DD1 random_JJ number_NN1 is_VBZ then_RT used_VVN to_TO select_VVI a_AT1 trade_NN1 from_II the_AT specified_JJ listing_NN1 ._. 
The_AT Monte_NP1 Carlo_NP1 simulation_NN1 can_VM reveal_VVI whether_CSW the_AT account_NN1 would_VM have_VHI experienced_VVN a_AT1 deeper_JJR drawdown_NN1 and_CC what_DDQ the_AT actual_JJ expected_JJ drawdown_NN1 could_VM be_VBI in_II real_JJ trading_NN1 ._. 
If_CS the_AT Monte_NP1 Carlo_NP1 drawdown_NN1 is_VBZ significantly_RR different_JJ to_II that_DD1 obtained_VVN from_II the_AT simulation_NN1 model_NN1 this_DD1 can_VM also_RR indicate_VVI that_CST the_AT system_NN1 has_VHZ been_VBN overoptimized_VVN ._. 
A_AT1 limitation_NN1 of_IO this_DD1 method_NN1 is_VBZ that_CST the_AT random_JJ number_NN1 generator_NN1 assumes_VVZ that_DD1 trade_NN1 returns_NN2 are_VBR independent_JJ of_IO each_PPX221 other_PPX222 ._. 
It_PPH1 is_VBZ therefore_RR a_AT1 good_JJ idea_NN1 to_TO check_VVI for_IF autocorrelation_NN1 of_IO returns_NN2 before_II running_VVG a_AT1 Monte_NP1 Carlo_NP1 simulation_NN1 ._. 
Low_JJ number_NN1 of_IO inputs_NN2 :_: By_II increasing_VVG the_AT number_NN1 of_IO inputs_NN2 to_II extremely_RR high_JJ levels_NN2 ,_, you_PPY could_VM design_VVI a_AT1 system_NN1 that_CST fits_VVZ almost_RR perfectly_RR around_II a_AT1 price_NN1 series_NN ._. 
This_DD1 system_NN1 ,_, however_RR ,_, will_VM almost_RR certainly_RR fail_VVI when_CS confronted_VVN with_IW real_JJ time_NNT1 data_NN that_CST it_PPH1 has_VHZ not_XX seen_VVN before_RT ._. 
It_PPH1 is_VBZ therefore_RR a_AT1 good_JJ idea_NN1 to_TO keep_VVI the_AT number_NN1 of_IO parameters_NN2 in_II the_AT model_NN1 relatively_RR low_JJ ,_, because_CS the_AT larger_JJR the_AT number_NN1 of_IO variables_NN2 in_II the_AT system_NN1 ,_, the_AT easier_JJR it_PPH1 is_VBZ to_TO create_VVI an_AT1 overoptimized_JJ system_NN1 that_CST will_VM underperform_VVI in_II real_JJ time_NNT1 ._. 
Diversification_NN1 with_IW intermarket_NN1 indicators_NN2 :_: Last_MD but_CCB not_XX least_RRT ,_, the_AT key_NN1 to_II robust_JJ system_NN1 development_NN1 is_VBZ to_TO select_VVI non-correlated_JJ inputs_NN2 that_CST have_VH0 predictive_JJ ability_NN1 ._. 
Using_VVG a_AT1 superfluous_JJ number_NN1 of_IO moving_VVG averages_NN2 and_CC oscillators_NN2 that_CST are_VBR derived_VVN from_II the_AT same_DA price_NN1 series_NN wo_VM n't_XX create_VVI a_AT1 profitable_JJ trading_NN1 strategy_NN1 ._. 
In_II fact_NN1 ,_, it_PPH1 will_VM most_RGT likely_RR fail_VVI to_TO work_VVI in_II real_JJ time_NNT1 ._. 
Rather_II21 than_II22 using_VVG standard_NN1 ,_, price_NN1 derived_VVD indicators_NN2 as_CSA your_APPGE only_JJ input_NN1 ,_, try_VV0 to_TO ascertain_VVI what_DDQ other_JJ markets_NN2 are_VBR related_VVN to_II the_AT market_NN1 being_VBG traded_VVN and_CC use_VV0 the_AT intermarket_NN1 relationship_NN1 as_II a_AT1 filter_NN1 or_CC replacement_NN1 of_IO a_AT1 redundant_JJ classic_JJ indicator_NN1 ._. 
One_MC1 interesting_JJ advantage_NN1 of_IO intermarket_NN1 analysis_NN1 is_VBZ that_CST it_PPH1 relies_VVZ on_II at_RR21 least_RR22 two_MC different_JJ data_NN sources_NN2 or_CC price_NN1 series_NN ._. 
Most_DAT single_JJ market_NN1 analysis_NN1 tools_NN2 are_VBR derived_VVN from_II the_AT same_DA price_NN1 data_NN ._. 
A_AT1 price_NN1 shock_NN1 or_CC bad_JJ data_NN would_VM affect_VVI all_DB the_AT indicators_NN2 or_CC tools_NN2 based_VVN on_II those_DD2 data_NN ._. 
By_II using_VVG two_MC different_JJ data_NN sources_NN2 ,_, you_PPY can_VM help_VVI to_TO insulate_VVI yourself_PPX1 against_II such_DA an_AT1 event_NN1 ._. 
The_AT use_NN1 of_IO intermarket_NN1 relationships_NN2 in_II developing_JJ trading_NN1 systems_NN2 is_VBZ explained_VVN extensively_RR in_II the_AT following_JJ chapters_NN2 ._. 
By_II using_VVG some_DD of_IO the_AT above_JJ techniques_NN2 ,_, I_PPIS1 am_VBM convinced_JJ that_CST anyone_PN1 can_VM ,_, at_II the_AT very_RG least_RRT ,_, avoid_VV0 trading_VVG with_IW systems_NN2 of_IO absolutely_RR no_AT value_NN1 ._. 
Considering_CS the_AT amount_NN1 of_IO money_NN1 at_II stake_NN1 in_II these_DD2 highly_RR leveraged_JJ and_CC volatile_JJ markets_NN2 ,_, it_PPH1 's_VBZ hard_JJ to_TO believe_VVI anyone_PN1 can_VM justify_VVI trading_VVG a_AT1 system_NN1 without_IW first_MD performing_VVG some_DD type_NN1 of_IO validation_NN1 and_CC reliability_NN1 testing_NN1 ._. 
Markets_NN2 :_: The_AT choice_NN1 of_IO securities_NN2 used_VVN in_II back-testing_NN1 is_VBZ also_RR important_JJ ._. 
For_REX21 example_REX22 ,_, if_CS a_AT1 broad_JJ market_NN1 system_NN1 is_VBZ tested_VVN only_RR on_II oil_NN1 stocks_NN2 ,_, it_PPH1 may_VM fail_VVI to_TO do_VDI well_RR in_II different_JJ sectors_NN2 ._. 
As_II a_AT1 general_JJ rule_NN1 ,_, if_CS a_AT1 strategy_NN1 is_VBZ targeted_VVN towards_II a_AT1 specific_JJ sector_NN1 ,_, limit_VV0 the_AT stocks_NN2 to_II that_DD1 sector_NN1 ,_, but_CCB in_II all_DB other_JJ cases_NN2 ,_, maintain_VV0 a_AT1 large_JJ universe_NN1 for_IF testing_VVG purposes_NN2 ._. 
Software_NN1 inputs_NN2 :_: Back-testing_JJ customization_NN1 is_VBZ extremely_RR important_JJ ._. 
Many_DA2 back-testing_JJ applications_NN2 have_VH0 input_VVN for_IF commission_NN1 amounts_NN2 ,_, round_JJ (_( or_CC fractional_JJ )_) lot_NN1 sizes_NN2 ,_, tick_VV0 sizes_NN2 ,_, margin_NN1 requirements_NN2 ,_, interest_NN1 rates_NN2 ,_, slippage_VV0 assumptions_NN2 ,_, position-sizing_JJ rules_NN2 ,_, same-bar_JJ exit_NN1 rules_NN2 ,_, (_( trailing_VVG )_) stop_VV0 settings_NN2 and_CC much_RR more_RRR ._. 
To_TO get_VVI the_AT most_RGT accurate_JJ back-testing_JJ results_NN2 ,_, it_PPH1 is_VBZ important_JJ to_TO tune_VVI these_DD2 settings_NN2 to_TO mimic_VVI your_APPGE trading_NN1 style_NN1 and_CC the_AT broker_NN1 that_CST will_VM be_VBI used_VVN when_CS the_AT system_NN1 goes_VVZ live_JJ ._. 
Paper_NN1 trading_NN1 :_: The_AT market_NN1 is_VBZ dynamic_JJ and_CC strategies_NN2 that_CST performed_VVD well_RR in_II the_AT past_NN1 fail_VV0 to_TO do_VDI well_RR in_II the_AT present_NN1 because_CS ofchanging_VVG market_NN1 conditions_NN2 ._. 
Be_VB0 sure_JJ to_II paper_NN1 trade_NN1 a_AT1 system_NN1 to_TO ensure_VVI that_CST the_AT strategy_NN1 still_RR applies_VVZ in_II practice_NN1 ._. 
Data_NN quality_NN1 :_: Price_NP1 data_NN often_RR contain_VV0 errors_NN2 ,_, which_DDQ can_VM cause_VVI serious_JJ problems_NN2 ,_, Special_JJ care_NN1 should_VM be_VBI exercised_VVN when_CS handling_VVG missing_JJ values_NN2 as_CSA these_DD2 are_VBR usually_RR replaced_VVN by_II zeroes_NN2 ._. 
Sometimes_RT free_JJ internet_NN1 data_NN sources_NN2 ,_, such_II21 as_II22 Yahoo_NP1 Finance_NN1 ,_, do_VD0 not_XX adjust_VVI stock_NN1 prices_NN2 for_IF splits_NN2 ._. 
The_AT resulting_JJ gap_NN1 will_VM not_XX only_RR affect_VVI the_AT system_NN1 's_GE profitability_NN1 but_CCB will_VM also_RR bias_VVI the_AT system_NN1 to_TO select_VVI the_AT appropriate_JJ parameters_NN2 to_TO position_VVI itself_PPX1 on_II the_AT right_JJ side_NN1 of_IO the_AT market_NN1 when_CS the_AT price_NN1 jump_NN1 (_( or_CC split_NN1 )_) occurred_VVD ._. 
While_CS easily_RR spotted_VVN manually_RR ,_, data_NN errors_NN2 tend_VV0 to_II corrupt_JJ a_AT1 mechanical_JJ system_NN1 ifa_NN1 price_NN1 series_NN is_VBZ used_VVN for_IF testing_VVG without_IW careful_JJ data_NN preprocessing_VVG and_CC error_NN1 checking_VVG ._. 
There_EX is_VBZ an_AT1 option_NN1 in_II Equis_NP1 '_GE Downloader_NN1 to_TO test_VVI the_AT data_NN for_IF over_RG 100_MC different_JJ error_NN1 conditions_NN2 ._. 
The_AT most_RGT useful_JJ is_VBZ testing_VVG for_IF a_AT1 large_JJ change_NN1 in_II close_JJ ._. 
